Adaptive filtering of evoked potentials with radial-basis-function neural network prefilter.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Neuroscience

Background:

• Evoked potentials (EPs) are crucial physiological signals often obscured by significant background noise.
• Previous methods for EP extraction, such as adaptive signal enhancers (ASE), relied on ensemble- and moving window-averages for reference signals.
• Optimal performance of ASE is contingent upon the quality of its reference input signal.

Purpose of the Study:

• To develop a more effective method for generating reference input signals for adaptive signal enhancers (ASE) used in evoked potential (EP) extraction.
• To enhance the signal-to-noise ratio and accuracy of EP detection.
• To evaluate the performance of a novel RBFNN-based approach compared to existing methods.

Main Methods:

• A Gaussian radial basis function neural network (RBFNN) was employed to preprocess raw evoked potential (EP) signals.
• The RBFNN-generated signals served as reference inputs for an adaptive signal enhancer (ASE).
• The performance of the ASE with the RBFNN reference was compared against traditional averaging techniques.

Main Results:

• The Gaussian radial basis function neural network (RBFNN) effectively tracked signal variations in raw EP data due to its nonlinear activation functions.
• The adaptive signal enhancer (ASE) utilizing RBFNN-generated reference signals demonstrated superior performance in extracting EPs compared to previous methods.
• The enhanced EP signals exhibited improved clarity and accuracy.

Conclusions:

• The integration of a Gaussian radial basis function neural network (RBFNN) significantly enhances the efficacy of adaptive signal enhancers (ASE) for evoked potential (EP) signal processing.
• This novel approach offers a more robust and accurate method for extracting weak EP signals from noisy biological data.
• The findings suggest a promising advancement in neurophysiological signal analysis techniques.